0.25
0.5
0.75
1.25
1.5
1.75
2
Gambling pigeons: Primary rewards are not all that matter
Published on Jul 28, 20151891 Views
In nature, actions rarely lead to reward 100% of the time; most decisions involve an element of risk or uncertainty. The human decision-making literature is full of examples in which people
Related categories
Chapter list
Gambling pigeons: Primary rewards are not all that matter00:00
My goal00:34
Which option would you pick?01:59
Most people play it safe02:20
Now, which option would you pick?02:28
Most people take the chance02:43
“Reflection Effect” Kahneman & Tversky, 197902:48
What if outcomes are instead learned by reinforcement? - 103:27
What if outcomes are instead learned by reinforcement? - 204:06
What if outcomes are instead learned by reinforcement? - 304:08
What if outcomes are instead learned by reinforcement? - 404:12
What if outcomes are instead learned by reinforcement? - 504:15
Some doors lead to losses - 104:19
Some doors lead to losses - 204:25
Some doors lead to losses - 304:27
Some doors lead to losses - 404:28
Some trials only give one door to ensure exposure to all contingencies04:32
After learning, people choose between fixed and risky doors04:42
Yes… but preference is biased in the opposite way05:05
Experienced choices: Gamble more on gain trials05:18
Described and experience-based decisions engage different brain regions05:37
Whyare people more risk seeking for gains than for losses in experience-based choice? - 106:10
Extremes are overweighted in memory07:04
Whyare people more risk seeking for gains than for losses in experience-based choice? - 208:05
Consider a loss choice: -20 vs 0/-4008:37
Decision context matters09:30
Is this bias unique to humans?10:07
Pigeon version of the door task - 110:41
Pigeon version of the door task - 211:43
Pigeon version of the door task - 311:58
Pigeon version of the door task - 412:13
Pigeon version of the door task - 512:25
Pigeon version of the door task - 612:43
Pigeons and people readily learned to choose high-value over low-value options12:51
Pigeons, like people, were riskier for highvalue choices than for low-value choices13:16
So…13:53
But, what if biases are “costly”?14:39
Choice between 50% and 100% reinforcement - 115:34
Choice between 50% and 100% reinforcement - 216:22
Choice between 50% and 100% reinforcement - 317:13
Procedure17:28
Pigeons respond sensibly on unsignaled procedure17:46
But small change in procedure…18:01
Signals during delay lead pigeons to make bad choices!18:22
Suboptimal choice depends on contiguity between choice and signals - 119:09
Suboptimal choice depends on contiguity between choice and signals - 219:24
Suboptimal choice depends on contiguity between choice and signals - 319:43
Choice between: signaled 20% vs. unsignaled 50% food20:13
Other findings21:13
Laude et al., 201222:07
Gambling humans22:23
Students who gambled made more suboptimally choices23:35
SiGN Hypothesis23:52
Illustration of SiGN Hypothesis25:07
Monkeys also choose signals - 125:30
Monkeys also choose signals - 226:19
Monkeys also choose signals - 326:37
Behavioral choice data26:51
Midbrain Dopamine Response27:01
Modelling Suboptimal Choice with RL?27:36
Conclusions28:42
Take home messages and speculations29:29
Thanks to31:49